LALAL.AI vs. VocalRemover: Stem Splitting Quality Analysis
Users often ask us how LALAL.AI differs from other popular stem splitters, and whether it can split audio files into individual tracks better than a particular service. It's always possible to take the path of least resistance and just say that ours is the best of the lot and call it a day. But that isn't our approach.
Instead, we thoroughly analyze how each splitter handles a set of the same songs. The key findings and a summary of the results are published here, on our blog. They are accompanied by the files of the original songs and the full stems extracted by each service.
This allows anyone to explore the capabilities of splitters without spending a penny and use the materials provided to verify the objectivity of the analysis and draw their own conclusions.
We’ve previously posted the splitting analyses of Moises, iZotope, Spleeter, and PhonicMind. Now, by popular demand, it's time to compare LALAL.AI with VocalRemover.
What Is VocalRemover?
VocalRemover.org is an online service that uses artificial intelligence to isolate and remove vocals, instrumentals, drums, and bass. It allows you to upload audio files and download the extracted stems in MP3 and WAV formats and in lossy quality.
In addition to the main functionality, the service offers such features as pitch and tempo adjustment, file cutting, multiple file joining, voice recording, karaoke voiceover recording, as well as key and BPM finding.
VocalRemover operates on a freemium model. You can split one file per day for free which should not exceed 10 minutes in duration. To split more, you must purchase a monthly or annual subscription for approximately $12 and $65 respectively. The premium subscription gives you the right to upload files that total up to 500 minutes per day and split them faster than on the free plan.
What Is LALAL.AI?
LALAL.AI is a stem separation tool available as a web service and a desktop application. It uses Phoenix, the in-house developed artificial intelligence, to split files into 10 stems: vocals, instrumental, drums, bass, acoustic guitar, electric guitar, piano, synthesizer, string instruments and wind instruments.
It allows you to upload audio and video files in such formats as MP3, OGG, WAV, FLAC, AVI, MP4, MKV, AIFF, and AAC. The split stems are downloaded in the same format as the original file. The stem quality can be both lossy and lossless depending on the quality of the uploaded file.
Besides stem isolation and removal, LALAL.AI offers a voice cleaning service for canceling out noise and eliminating extraneous sounds (like background music, vocal plosives, microphone rumble, loud breathing, etc.) in voice and vocal recordings. Like the stem splitter, it supports audio and video formats.
Similar to VocalRemover, LALAL.AI offers a free trial that allows you to split a track of up to 10 minutes, after which you will need to upgrade to process more files. However, instead of a subscription, there are several one-time purchase options ranging from $15 to $300. All unlock the same set of premium features: 2GB per file upload, batch upload, and fast file processing. The only difference is the number of minutes you can split in total – it’s 90 minutes with the cheapest package and 5000 minutes with the most expensive one.
How We Compared the Two
Although VocalRemover and LALAL.AI can extract several different stems, in this analysis, we only compare how the services isolate vocal stems. There are a few reasons why:
- The vocal stem is the most popular and important for the vast majority of users across the board.
- Vocal extraction is a particularly difficult task. Just because it's done with state-of-the-art technology doesn't mean it's done right. How a service handles vocal isolation is considered an indicator of its sophistication and quality.
- Comparing each stem (drums, bass, and instrumental in addition to vocals) is possible, but it would make this already long article even longer and harder for readers to digest. If there is a demand, we are happy to do a separate analysis of drums and bass as well.
We uploaded a set of the same songs into VocalRemover and LALAL.AI, selected the vocal stem extraction, and used psychoacoustic methods (listened to the results made by both services) to determine how each of the services handled the splitting.
Key moments were cut out of the extracted stems and displayed below accompanied by commentary about the splitting quality. The cut-out excepts can be played right on the page.
The test songs and full versions of the extracted stems are also available below as downloadable files so that anyone can see that the results displayed here weren’t altered in favor of one or the other service. You are also free to use all the attached materials to perform the same test yourself.
Splitting Quality Analysis
The original unprocessed test songs and full stems can be played and downloaded by clicking on the song bars below.
🟡Song #1 – Full Audio & Stems
Song name: HumanFace – After Lunch
Except #1
In this section, you can hear gaps in the vocals of both LALAL.AI and VocalRemover, but LALAL.AI's are smoother than VocalRemover's. You can also hear hissing in the vocals in the VocalRemover excerpt.
Excerpt #2
This section serves as an excellent example of distortion (hissing) in the vocals of VocalRemover that is not present in the vocals of LALAL.AI.
🟡Song #2 – Full Audio & Stems
Song name: Maleibow — Some Places
Except #1
In this excerpt, both LALAL.AI and VocalRemover have guitar strumming in the vocals, but VocalRemover's is less noticeable.
Excerpt #2
Here you can hear the guitar bleeding into the vocals in both the LALAL.AI and VocalRemover excerpts, but LALAL.AI has a phaser effect, while VocalRemover has distortion (hissing).
🟡Song #3 – Full Audio & Stems
Song name: Graybo — Boys In Their Games
Here in VocalRemover's stem, the instrumental part bleeds into the vocals in some places, especially the electric guitar. Even during the silent parts, you can hear the instrumental oozing with a hiss, which is especially noticeable when listening in headphones.
🟡Song #4 – Full Audio & Stems
Yet another example of the phaser effect in the LALAL.AI vocal stem, while VocalRemover has a hiss instead.
Conclusion
In general, VocalRemover proved to be a worthy competitor to LALAL.AI, isolating vocals well, but not without errors. In the vocal stems extracted from each test song, the same problem was observed – distortion in the form of hissing, which seems to be there either to mask the effect of the phaser or the instrumental leakage, which sounds like crackling.
Even looking at the spectrograms of the stems on SoundCloud, it is obvious that in places where there should be no vocals, there are some extraneous sounds:
This is perhaps the most telling example of the noise in a place where there is supposed to be silence:
If you want to extract vocals from a song and not waste time and effort cleaning it of excess noise, LALAL.AI is a better choice. Of course, the final decision is all yours. Our goal is simply to demonstrate the splitting quality of the stem separation tools as objectively as possible.
We would also like to remind you that no service (yes, including ours) and no method of audio source separation, no matter how advanced, can provide 100% ideal results. Only the original stems in the possession of the recording engineer and performer are of perfect quality.
Check out our previous stem-splitting quality comparison articles: